A brain learns from a few experiences.So should a machine.
We are not chasing more data. We build AI that learns fast, the way the one system that truly acts intelligently in a messy, changing physical world does: the brain.
Senses everything. Reacts instantly.
Today's robot policies look at the scene, generate a whole chunk of actions, execute them blind, and only then look again. Touch, force and the other sensors mostly go unused.
Our models work differently. They sense everything, vision, position, touch, force, and every new signal can change the very next action. The loop is closed continuously, so the robot reacts the instant the world changes.
A handful of demos
A few demonstrations on your cell, not millions of samples.
Reacts at every step
Every sensor signal can change the next action. No chunk of actions executed blind.
Improves over time
Each attempt improves the next. Parts or light drift? Absorbed, not re-engineered.
Models that behave like real brains.
A brain predicts what will happen next and learns from the gap between what it expected and what it senses. That is how it picks up a skill from a few experiences and very little energy, where today's AI needs millions of samples and a data centre. For more than twenty years, at EPFL and through the Blue Brain Project, Henry Markram and his teams digitally reconstructed how real brains are built, neuron by neuron, synapse by synapse. Inait is the company that turns that science into working intelligence for machines. Watch Henry's talk at the World Economic Forum →
- •A scientific basis for physical AI. Brains learn from a few attempts, on almost no energy. Our models inherit that property.
- •Validated against real brains. Measured neuron by neuron against biological recordings, in more than 300 peer-reviewed papers.
- •Built at EPFL, ours exclusively. Hundreds of millions of public science went into the Blue Brain Project. Inait alone can build products on these models.
The Open Brain Institute is the non-profit that carries digital brain building forward for science: it hosts Virtual Labs where researchers build and simulate digital brain models to test hypotheses and explore neural circuits. Inait takes the same science and puts it to work in machines.
A digital brain learns pong in eleven games.
Pong is the classic video game: a ball, two paddles, and you have to hit it back. A digital brain of 500 neurons, with no prior knowledge of the game, learns to win on a laptop CPU in under five minutes.
The same digital brain can learn to play in different styles.
What pong has to do with your factory.
Pong is the proof that a digital brain learns fast, from a few experiences, on ordinary hardware. We are bringing the same fast learning to industrial robots, where a new task has to be learned from a few demonstrations, on the cell that already exists. See the first results →
Brains for machines.
The digital brain is Inait's crown jewel, and just the core of a broader platform of proprietary, brain-like AI tools, productized as iAM Robotics. Three layers turn the science into something a robot can run.
The recipe to copy a brain digitally.
Reconstructed neuron by neuron and synapse by synapse, and validated against biological recordings in more than 300 peer-reviewed papers.
Makes the simulation run fast, with real inputs and outputs.
Turns the simulation into a high-performance spiking neural network, connected to sensors and actuators, with learning that continues while it runs.
Brain-like AI tools for real industrial robots, ready for the shop floor.
- Real-time learning across many sensors. Fuses many sensor streams into one live software model.
- Integrates into your stack. Microsoft's robotics toolchain, leading robot OS providers.
- Runs on any hardware. GPU or CPU, on-premise or cloud.
Any machine that has to sense, decide and act in a changing world.
We started with advanced manufacturing on purpose. It has a large number of well-defined, hard tasks, and a factory does not forgive weak claims: either it works 99% of the time on the line, or it does not. We are now moving to more demanding industrial tasks: complex kitting, assembly, handling cables in a data centre. Further out, the same Inait intelligence layer extends to logistics, healthcare and humanoids, for the most challenging and constantly changing environments.